Existing metadata term recommendation algorithms for Linked Open Data (LOD) mainly rely on the frequency of metadata terms and do not sufficiently consider the domains in which metadata terms should be applied. As a result, users may not receive recommendations suited to their specific domains, limiting term reuse. This study proposes a new metadata term recommendation algorithm that incorporates both overall usage trends and domain-specific usage patterns. By reflecting the differences in term applicability, the proposed method enables more appropriate recommendations. Evaluation using domain-specific datasets demonstrates its effectiveness. Future work includes further improvements by considering vocabulary structures and dataset schema information.